Gearbox Fault Diagnosis Method Based on Multidomain Information Fusion.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Traditional gearbox fault diagnosis relies on manual experience, limiting accuracy and efficiency.
• Developing automated and accurate fault diagnosis methods is crucial for industrial machinery.
• Existing methods often struggle with complex vibration patterns and noise.

Purpose of the Study:

• To propose a gearbox fault diagnosis method based on multidomain information fusion.
• To enhance the accuracy and reliability of gearbox fault recognition.
• To overcome the limitations of traditional experience-based diagnosis methods.

Main Methods:

• Constructed an experimental platform with a JZQ250 fixed-axis gearbox and acceleration sensor.
• Preprocessed vibration signals using Singular Value Decomposition (SVD) and Short-Time Fourier Transform (STFT) for time-frequency analysis.
• Developed a multidomain information fusion Convolutional Neural Network (CNN) model with parallel 1DCNN and 2DCNN channels, fused features, and Support Vector Machine (SVM) classification.

Main Results:

• The proposed multidomain information fusion CNN model achieved the highest fault recognition accuracy of 98.08%.
• Experimental verification demonstrated superior performance compared to FFT-2DCNN, 1DCNN-SVM, and 2DCNN-SVM methods.
• t-SNE visualization confirmed effective feature extraction and classification.

Conclusions:

• The multidomain information fusion method significantly improves gearbox fault diagnosis accuracy.
• The integrated approach of signal processing, CNN, and SVM offers a robust solution for automated gearbox health monitoring.
• This study provides a promising direction for advancing intelligent fault diagnosis in mechanical systems.